thomassavary08/lyapynov

Python package to compute Lyapunov exponents, covariant Lyapunov vectors (CLV) and adjoints of a dynamical systems.

Lyapynov

Lyapynov is a Python library to compute Lyapunov exponents, covariant Lyapunov vectors (CLV) and their adjoints for a dynamical system. The results/algorithms used are taken from P. Kuptsov's paper on covariant Lyapunov vectors.

Installation

Use the package manager pip to install Lyapynov.

pip install lyapynov

Usage

First, one needs to define the system to study (which can be discrete or continuous) using the ContinuousDS or DiscreteDS methods. These methods take the following parameters as input:

• the initial conditions $x_{0}$ and $t_{0}$ $(x(t_{0}) = x_{0})$.
• the function $f$ describing the dynamical system:

$$\left\{ \begin{array}{ll} \dot{x} = f(x,t) \\\ x_{n+1} = f(x_{n},n) \end{array} \right.$$

• the jacobian of $f$ with respect to $x$ or $x_{n}$:

$$\left\{ \begin{array}{ll} J(x,t) = \displaystyle \frac{\partial f}{\partial x}(x,t) \\\ ~ \\\ J(x_{n},n) = \displaystyle \frac{\partial f}{\partial x_{n}}(x_{n},n) \end{array} \right.$$

import lyapynov

# Continous dynamical system
continuous_system = lyapynov.ContinuousDS(x0, t0, f, jac, dt)

# Discrete dynamical system
discrete_system = lyapynov.DiscreteDS(x0, t0, f, jac)

Once the dynamic system has been defined, the following functions can be used:

• Maximum Lyapunov exponents (MLE) to compute the maximum Lyapunov exponent $\lambda_{1}$.

def mLCE(system : DynamicalSystem, n_forward : int, n_compute : int, keep : bool):
    '''
    Compute the maximal 1-LCE.
        Parameters:
            system (DynamicalSystem): Dynamical system for which we want to compute the mLCE.
            n_forward (int): Number of steps before starting the mLCE computation. 
            n_compute (int): Number of steps to compute the mLCE, can be adjusted using keep_evolution.
            keep (bool): If True return a numpy array of dimension (n_compute,) containing the evolution of mLCE.
        Returns:
            mLCE (float): Maximum 1-LCE.
            history (numpy.ndarray): Evolution of mLCE during the computation.
    '''

• Lyapunov characteristic exponents (LCE) to compute the $p$ first Lyapunov exponents $(\lambda_{1}, \cdots, \lambda_{p})$.

def LCE(system : DynamicalSystem, p : int, n_forward : int, n_compute : int, keep : bool):
    '''
    Compute LCE.
        Parameters:
            system (DynamicalSystem): Dynamical system for which we want to compute the LCE.
            p (int): Number of LCE to compute.
            n_forward (int): Number of steps before starting the LCE computation. 
            n_compute (int): Number of steps to compute the LCE, can be adjusted using keep_evolution.
            keep (bool): If True return a numpy array of dimension (n_compute,p) containing the evolution of LCE.
        Returns:
            LCE (numpy.ndarray): Lyapunov Charateristic Exponents.
            history (numpy.ndarray): Evolution of LCE during the computation.
    '''

• Covariant Lyapunov vectors (CLV) to compute covariant Lyapunov vectors $\Gamma(t) = [\gamma_{1}(t), \cdots, \gamma_{m}(t)]$.

def CLV(system : DynamicalSystem, p : int, n_forward : int, n_A : int, n_B : int, n_C : int, traj : bool, check = False):
    '''
    Compute CLV.
        Parameters:
            system (DynamicalSystem): Dynamical system for which we want to compute the mLCE.
            p (int): Number of CLV to compute.
            n_forward (int): Number of steps before starting the CLV computation. 
            n_A (int): Number of steps for the orthogonal matrice Q to converge to BLV.
            n_B (int): Number of time steps for which Phi and R matrices are stored and for which CLV are computed.
            n_C (int): Number of steps for which R matrices are stored in order to converge A to A-. 
            traj (bool): If True return a numpy array of dimension (n_B,system.dim) containing system's trajectory at the times CLV are computed.
        Returns:
            CLV (List): List of numpy.array containing CLV computed during n_B time steps.
            history (numpy.ndarray): Trajectory of the system during the computation of CLV.
    '''

• Adjoint covariant vectors (ADJ) to compute the adjoints of CLV $\Theta(t) = [\theta_{1}(t), \cdots, \theta_{m}(t)]$ such that $\Gamma(t)^{T} \Theta(t) = D(t)$ .

def ADJ(CLV : list):
    '''
    Compute adjoints vectors of CLV.
        Parameters:
            CLV (list): List of np.ndarray containing CLV at each time step: [CLV(t1), ...,CLV(tn)].
        Returns:
            ADJ (List): List of numpy.array containing adjoints of CLV at each time step (each column corresponds to an adjoint).
    '''

An example of using the package is given in the notebook Example.ipynb.